Title:Pain Detection with fNIRS-Measured Brain Signals: A Personalized Machine Learning Approach Using the Wavelet Transform and Bayesian Hierarchical Modeling with Dirichlet Process Priors

Abstract:Currently self-report pain ratings are the gold standard in clinical pain assessment. However, the development of objective automatic measures of pain could substantially aid pain diagnosis and therapy. Recent neuroimaging studies have shown the potential of functional near-infrared spectroscopy (fNIRS) for pain detection. This is a brain-imaging technique that provides non-invasive, long-term measurements of cortical hemoglobin concentration changes. In this study, we focused on fNIRS signals acquired exclusively from the prefrontal cortex, which can be accessed unobtrusively, and derived an algorithm for the detection of the presence of pain using Bayesian hierarchical modelling with wavelet features. This approach allows personalization of the inference process by accounting for inter-participant variability in pain responses. Our work highlights the importance of adopting a personalized approach and supports the use of fNIRS for pain assessment.